Google launches MedLM generative AI models fine-tuned for healthcare

Since the launch of ChatGPT by OpenAI, generative AI has taken the world by storm. Every industry is now looking for ways in which gen AI can simplify their work and reduce costs. And healthcare is not behind.

Generative AI in Healthcare market is expected to grow at a CAGR of ~36% and reach USD 25 Billion by 2032 and Google is all set to make most of it.

Last week, the tech giant unveiled MedLM, its new suite of generative artificial intelligence (AI) models fine-tuned for the healthcare industry.

From improving patient care to streamlining administrative tasks, MedLM holds the potential to revolutionise healthcare. Let’s find out how.

An overview of Google’s MedLM

Last year, Google made headlines with its large language model (LLM) Med-Palm, which cleared the US medical licensing exam with >60% accuracy. This March, Google announced Med-PaLM 2, an updated version, with 85% accuracy.

Now, Google has further built on Med-PaLM 2 to create MedLM, a family of LLMs specifically catered to the healthcare industry.

Two models of MedLM

Combining the power of generative AI with LLMs, MedLM currently consists of two models to meet the different healthcare needs of different organisations.

The first MedLM model is larger and designed for complex tasks like summarising doctor-patient conversations into medical notes.

The second is a medium model and is best for fine-tuning the tool and scaling it across tasks. For example, simplifying searching through medications and other admin tasks in healthcare.

MedLM use cases

Google partnered with leading companies in the healthcare space like HCA Healthcare, Accenture, Deloitte and BenchSci to test MedLM and fine-tune its capabilities. Here are some use cases of MedLM as showcased by Google in its blog post:

Medical documentation with HCA Healthcare

HCA Healthcare piloted MedLM with Augmedix, an app using natural language processing to translate patient-doctor conversations into medical notes. MedLM improved the quality of these medical notes drafts, which the physicians could review and transfer directly into the patient’s EHR.

Improving patient outcomes with Accenture

Accenture teamed up Google’s MedLM with its Solutions.AI for Processing for Health to speed up the processing of clinical documents, enrollment forms, medical claims and more. This assists clinicians to make faster and more informed decisions while freeing up more time and resources for patient care.

Optimising provider search with Deloitte

Deloitte combined MedLM with its patient-provider matching services for insurance plans. This allows its contact centre agents to match a patient with the best-fit provider based on the patient’s plan, health condition, medications and prior appointment history.

Streamlining drug research and development with BenchSci

BenchSci integrated MedLM with its ASCEND platform to improve the speed and quality of pre-clinical research and development. MedLM further enhanced ASCEND’s accuracy, precision and reliability in identifying biomarkers to accelerate drug development.

Availability of MedLM

Currently, MedLM is available to US-based Google Cloud customers (allow listed availability) through the Vertex AI platform. It is also available for preview in certain other countries. 

Google’s future plans for MedLM

The tech giant plans to bring Gemini-based models into MedLM in the coming months. Gemini is Google’s flagship gen AI model positioned as a tough competition to ChatGPT by OpenAI. By adding Gemini-based models, Google plans to expand the capabilities of MedLM.

We may see Google adding domain-specific models which are “smaller, cheaper, faster, better,”

– said Greg Corrado, head of Google’s health AI, in an interview with CNBC.

Potential of LLMs like MedLM in healthcare

AI-powered LLM, like MedLM, has immense potential in healthcare. From streamlining processes and assisting healthcare professionals in decision-making to accelerating medical research and improving patient outcomes.

However, AI has its own limitations. It cannot understand user intent or give accurate answers when a patient doesn’t know the right word or spelling. It also has a risk of bias depending on the data used to train it.

Therefore, it’s important to understand that LLM can only be used as a tool to assist a provider and not as a substitute for diagnosis from an expert.

That being said, research around MedLM is still ongoing. With time, we may see further fine-tuning and development so that it can better assist healthcare professionals and staff.